Loads of doubles

[rombout]

In last weeks i had a lot of matches with almost as many doubles as "normal" dice cimbinations (on average it should be 1:6)
Of course this could happen sometimes but now it's happening a lot. So much that it takes the fun out of the game (it's not that I only lose because of this)
Was something changed in the algorithm? Can that be undone?
I would like to have the fun back in.

[euklid314]

I looked at your last 10 games and do not know what you are talking about. There is not a single game where you have a 1:1 ratio, not even close. 4 out of 18, 2 out of 39 (your opponent), 6 out of 39 (yourself), and so on. I did not calculate the average over those 10 games but just a fast glimpse at the numbers (you get a statistic in the end of each game) shows nothing of a problem.

Be assured the algorithm was not changed... Have fun!

[gammon_rant]

You need to play more games and at least 1000 dice throws if you want to have a decent sample to analyse. Under 1000 dice there is too much volatility as you can see in this picture:



This is from a book I wrote (in french), backgammon masterclass.

[Ze Monstah]

[Romain672]

(edit: this post is completely wrong: i multiply the number of doubles by 2)

Let's focus on double since it's the most used factor:
[numberofthegame]: [number of doubles for the yellow player]y [number of doubles for the red player]r #[number of turns before the end of the game] ([color of Ze Monstah]){[estimated number of rolls for each player]}[sum of doubles]
163852413: 2y 4r #93 (r){30.6}6
163842398: 3y 4r #90 (y){29.1}7
163760484: 5y 2r #78 (r){24.9}7
163750180: 4y 5r #102 (y){32.6}9
163749910: 2y 3r #82 (y){27.1}5
163733497: 6y 4r #79 (r){23.9}10
163731807: 1y 2r #40 (y){13}3
163737793: 3y 3r #77 (y){24.9}6
163730921: 3y 1r #56 (y){18.3}4
163729488: 5y 4r #89 (r){28}9

So here the probability to do that number of double or more with this number of rolls:
25, 10, 4, 4, 29, 0 (0.3), 16, 11, 17, 1

Since the last time I checked your profile, you did exactly 170 games.
That should make any unlikely event in 10 games about ~17x more likely (or ~x160 if you count the differents suits of 10).

But god, that's a pretty weird serie.
I tried to reproduce it by a random generation, after 4128 trys, I decided to speed up my random generation per 10, and after 147600 more trys, I gave up.

Am I missing something? Did I make a mistake with the games? If not, I will goes into Ze Monstah's camp soon.

And I regulary talk with Ze Monstah on the forum, which mean it's one of the most valid suits possible, and I took one of the four 'purest' things you can check (ie: number of your double, number of opponent double, number of double, and streak of win/loss).

[N_Faker]

Your estimated number of rolls is all over the place.

I didn't bother checking who was red or yellow so the numbers may be swapped. You miscounted the number of doubles for some games too, those are in red.

Have fun.

163852413: 2y 4r 27
163842398: 3y 4r 24
163760484: 5y 2r 25.5
163750180: 5y 5r 31.5
163749910: 2y 3r 22
163733497: 6y 4r 16.5
163731807: 1y 2r 11
163737793: 3y 3r 18.5
163730921: 3y 2r 17
163729488: 5y 6r 19

[dschingis27]

I absolutely cannot see how the series of doubles is weird.?! Could you explain what the "estimated number of rolls for each player" actually is? Since there is an actual number of rolls, it's not clear to me on what premise you calculate an estimated number of rolls.

So here the probability to do that number of double or more with this number of rolls:
25, 10, 4, 4, 29, 0 (0.3), 16, 11, 17, 1

I would be interested to know how these probablities were calculated. I feel pretty confident that these are incorrect.

[dschingis27]

I don't have the whole programming IT algorithm, only a part.

Please... someone please take a look at the algorithm used for this game...

An IT individual, an impartial programmer.

The IT guy looking into this has already happened: https://boardgamearena.com/forum/viewto ... 1&start=10

You can argue that these are "just" simulations but I would say that this is the best we can do. Not even a slight indication of a problem has been found.

Regarding the code:

// Roll dices
$dice1_value = bga_rand(1, 6);
$dice2_value = bga_rand(1, 6);

bga_rand() function is defined like this:

function bga_rand( $min=null, $max=null )
{
if( $min === null && $max === null )
return random_int( 0, getrandmax() ); // For compatibility
else
return random_int( $min, $max );
}

This is the damn complete code. And there is absolutely no problem to it. This was discussed already here https://boardgamearena.com/forum/viewto ... d&start=30 The function random_int() was not created by BGA and it is a validated PHP function also used by other websites.

There are two things left to consider. A) Theoretically, there could be some problem not detected by the statistics from the simulation I linked above. That would then be most likely an extremely weird hardware issue with the servers. But if so, these problems would be extremely hard to detect. B) Imagnine there is a problem inherent in the function random_int() itself (that would then also concern many other systems outside BGA). This is even more unlikely than A) and would also require extreme effort to detect.

It is unfortunate indeed, that we cannot use webscraping or whatever to produce statistics of a big number of actually played games on BGA.
But all in all, based on the facts and stats we know, it is just super extreme unlikely that there is any issue with the RNG.

[Ze Monstah]

[whatshisbucket]

Your paragraphs and paragraphs of complaints without evidence are completely worthless. If you want to prove that something is wrong, get off your ass and collect some data. Frankly, the way for you to collect data that is simultaneously the least biased and requires the least effort for you is to pick some fixed number of games beforehand (say, 200) and collect data on the very next games you play (this way, if something tangible and 1 in a million happens, it is more likely that it is because there is a problem with the RNG and not because one lucky user had it happen to them and came to report it after the fact). I'm sure the effort required to collect that much data is less than the effort you've already wasted complaining.